Jakarta, INTI – Amazon Web Services (AWS) is once again asserting its presence in the artificial intelligence (AI) industry by unveiling its latest chip, Trainium 3, at AWS re:Invent 2025 in Las Vegas on Tuesday, December 2, 2025. The chip is claimed to boost AI model training and inference performance up to four times faster than the previous generation, while consuming up to 40 percent less energy. This launch comes as a solution for companies that require large-scale AI computing with more cost- and energy-efficient operations. Trainium 3 has already begun to be adopted by major companies, including Anthropic and the Japanese AI startup, Karakuri.
High Performance with Large Memory Capacity
Trainium 3 is built using 3-nanometer fabrication technology, one of the most advanced chip manufacturing processes available today. AWS states that the chip not only accelerates AI model training but also increases memory capacity up to four times compared to Trainium 2. A single Trainium 3 server can house up to 144 chips, allowing configurations of thousands of servers to form clusters containing more than one million chips, delivering computing power up to 362 petaflops with an aggregate bandwidth exceeding 700 terabits per second.
Energy Efficiency and Reduced Inference Costs
Another key advantage of Trainium 3 is energy efficiency. With 40 percent lower power consumption, the chip helps companies reduce data center operational costs while significantly lowering AI inference costs. Several companies, including Anthropic and Karakuri, have reported substantial reductions in the operating costs of their AI models after adopting Trainium 3.
Trainium 4 Roadmap and NVLink Fusion Support
AWS also revealed the roadmap for the next generation, Trainium 4, scheduled to debut at AWS re:Invent 2026. Trainium 4 is expected to deliver up to six times higher performance, larger memory capacity, and four times the memory bandwidth. Notably, the chip will support Nvidia’s NVLink Fusion, enabling seamless integration between AWS Trainium chips and Nvidia GPUs within a single server. This combination allows companies to leverage the performance of Nvidia GPUs while benefiting from AWS’s cost efficiency and energy savings.
Conclusion
The launch of Trainium 3 reinforces AWS’s position as a major player in the global AI chip market, offering fast, energy-efficient, and cost-effective AI computing solutions. With the Trainium 4 roadmap and NVLink Fusion support, AWS opens new opportunities for companies to develop large-scale AI models more efficiently without compromising performance.
Read More:The First AI Chef in Dubai Combines Technology and Gastronomy